Dorota Witkowska, Ph.D., Dept. of Econometrics & Statistics, Warsaw University of Life Sciences, ul. Nowoursynowska 166, Warszawa, 02-787, Poland
Participation Units of the Investment Fund Forecasting: Neural Network Application
Investment funds have been developing since 80’s of 20th century. The first investment fund in Poland was founded in 1992, and the significant development has been observed since that time. In 2004 an essential change of legal rules concerning organization and functioning of investment funds in Poland took place. This change was caused by Poland’s accession to the European Union. Due to these regulations investment funds in Poland operate as: open – end funds, specialistic open – end funds and closed – end funds.
Participation units of the open – end investment fund inform about the value of the investment fund assets. Thus forecast of the participation units is an important information for the investor.
Artificial neural networks are non-linear models that can be trained to extract hidden structures and relationships that govern the data. They can be used for analyzing relations among economic and financial phenomena. Several authors have examined the application of ANN to financial markets, where the non-linear properties of financial data provide many difficulties for traditional methods of analysis (Omerod, Taylor, and Walker (1991); Grudnitski and Osburn (1993); Altman, Marco, and Varetto (1994); Kaastra and Boyd (1995) as well as Hawley, Johnson, and Raina (1990); White (1988)). Hertz, Krogh, and Palmer (1991) offer a comprehensive view of neural networks and issues of their comparison to statistics. Hinton (1992) investigates the statistical aspects of ANN. Weiss and Kulikowski (1991) offer an account of the classification methods of many different neural and statistical models. Yoon and Swales (1990) compare ANN to discriminant analysis with respect to prediction of stock price performance and find that the neural network is superior to discriminant analysis in its predictions.
The aim of the research is to construct the neural model to forecast the unit of the participation in investment fund value. Research is conducted using daily data of the participation unit values of the investment fund SKARBIEC AKCJA in the period from 2.01.2003 to 28.02 2006. The accuracy of forecasts, generated by multilayer perceptron, is evaluated in terms of root squared error and correlation coefficient between actual and predicted values.